Automated Segmentation of Brain FDG PET Images Using a Gaussian Mixture Model for the Measurement of Aging Effects

  • MASUDA Yasushi
    Graduate School of Engineering Science, Osaka University
  • KIMURA Yuichi
    Positron Medical Center, Tokyo Metropolitan Institute of Gerontology
  • NAGANAWA Mika
    Graduate School of Information Science, Nara Institute of Science and Technology
  • CHIHARA Kunihiro
    Graduate School of Information Science, Nara Institute of Science and Technology
  • OSHIRO Osamu
    Graduate School of Engineering Science, Osaka University

Bibliographic Information

Other Title
  • 加齢傾向検討のためのガウス混合モデルによる脳PET画像の自動セグメンテーション
  • カレイケイコウ ケントウ ノ タメ ノ ガウス コンゴウ モデル ニ ヨル ノウ PET ガゾウ ノ ジドウ セグメンテーション

Search this article

Abstract

This paper presents a method of volume segmentation for brain fluorodeoxyglucosepositron emission tomography (FDG-PET) images using a Gaussian mixture model (GMM). In FDG-PET diagnosis, an abnormal decrease of regional cerebral glucose metabolism (CMRGlc) should be discriminated from one caused by normal brain atrophy as the effect of aging. To achieve it without any incorporation of a structural image such as magnetic resonance imaging (MRI), a volumetric study on FDG-PET image sequences should be performed; thus, a robust method for volume segmentation with CMRGlc is required. The proposed method assumes that the probabilistic distribution of CMRGlc has three groups; originated from the gray matter, white matter, and cerebrospinal fluid space. Two consecutive GMM runs successfully clustered FDG-PET images into three clusters resulting in robust volume segmentation without any heuristic operation such as thresholding. The proposed method was applied to FDG-PET images from 71 healthy volunteers, and a significant increase in the low CMRGlc region on age was observed. We conclude that successive GMM is useful for the segmentation of PET FDG images.

Journal

References(16)*help

See more

Details 詳細情報について

Report a problem

Back to top